feat: add prometheus metrics network_version and actor_version

CHANGELOG.md

@@ -30,7 +30,10 @@
 ### Added
 
 - [#6057](https://github.com/ChainSafe/forest/issues/6057) Added `--no-progress-timeout` to `forest-cli f3 ready` subcommand to exit when F3 is stuck for the given timeout.
-- [#6000](https://github.com/ChainSafe/forest/pull/6000) Add support for the `Filecoin.StateDecodeParams` API methods to enable decoding actors method params.
+
+- [#6000](https://github.com/ChainSafe/forest/pull/6000) Added support for the `Filecoin.StateDecodeParams` API methods to enable decoding actors method params.
+
+- [#6079](https://github.com/ChainSafe/forest/pull/6079) Added prometheus metrics `network_version`, `network_version_revision` and `actor_version`.
 
 - [#6068](https://github.com/ChainSafe/forest/issues/6068) Added `--index-backfill-epochs` to `forest-tool api serve`.
 

docs/docs/users/reference/metrics.md

@@ -17,6 +17,9 @@ title: Metrics
 | `full_peers`                   | Gauge     | Count   | Number of healthy peers recognized by the node                                               |
 | `bad_peers`                    | Gauge     | Count   | Number of bad peers recognized by the node                                                   |
 | `expected_network_height`      | Gauge     | Count   | The expected network height based on the current time and the genesis block time             |
+| `network_version`              | Gauge     | Count   | Network version of the current chain head                                                    |
+| `network_version_revision`     | Gauge     | Count   | Network version revision of the current chain head                                           |
+| `actor_version`                | Gauge     | Count   | Actor version of the current chain head                                                      |
 | `forest_db_size`               | Gauge     | Bytes   | Size of Forest database in bytes                                                             |
 | `bitswap_message_count`        | Counter   | Count   | Number of `bitswap` messages. Indexed by `type`                                              |
 | `bitswap_container_capacities` | Gauge     | Count   | Capacity for each `bitswap` container. Indexed by `type`                                     |
@@ -288,6 +291,33 @@ expected_network_height 2519530
 ```
 </details>
 
+<details>
+  <summary>Example `network_version` output</summary>
+```
+# HELP network_version Network version of the current chain head
+# TYPE network_version gauge
+network_version 27
+```
+</details>
+
+<details>
+  <summary>Example `network_version_revision` output</summary>
+```
+# HELP network_version_revision Network version revision of the current chain head
+# TYPE network_version_revision gauge
+network_version_revision 0
+```
+</details>
+
+<details>
+  <summary>Example `actor_version` output</summary>
+```
+# HELP actor_version Actor version of the current chain head
+# TYPE actor_version gauge
+actor_version 17
+```
+</details>
+
 <details>
   <summary>Example `build_info` output</summary>
 ```

src/chain/store/chain_store.rs

@@ -6,6 +6,7 @@ use super::{
     index::{ChainIndex, ResolveNullTipset},
     tipset_tracker::TipsetTracker,
 };
+use crate::db::{EthMappingsStore, EthMappingsStoreExt, IndicesStore, IndicesStoreExt};
 use crate::interpreter::{BlockMessages, VMTrace};
 use crate::libp2p_bitswap::{BitswapStoreRead, BitswapStoreReadWrite};
 use crate::message::{ChainMessage, Message as MessageTrait, SignedMessage};
@@ -22,10 +23,6 @@ use crate::{
     blocks::{CachingBlockHeader, Tipset, TipsetKey, TxMeta},
     db::HeaviestTipsetKeyProvider,
 };
-use crate::{
-    chain_sync::metrics,
-    db::{EthMappingsStore, EthMappingsStoreExt, IndicesStore, IndicesStoreExt},
-};
 use crate::{fil_cns, utils::cache::SizeTrackingLruCache};
 use ahash::{HashMap, HashMapExt, HashSet};
 use anyhow::Context as _;
@@ -146,7 +143,6 @@ where
 
     /// Sets heaviest tipset
     pub fn set_heaviest_tipset(&self, ts: Arc<Tipset>) -> Result<(), Error> {
-        metrics::HEAD_EPOCH.set(ts.epoch());
         self.heaviest_tipset_key_provider
             .set_heaviest_tipset_key(ts.key())?;
         if self.publisher.send(HeadChange::Apply(ts)).is_err() {

src/chain_sync/metrics.rs

@@ -3,7 +3,7 @@
 
 use prometheus_client::{
     encoding::{EncodeLabelKey, EncodeLabelSet, EncodeLabelValue, LabelSetEncoder},
-    metrics::{counter::Counter, family::Family, gauge::Gauge, histogram::Histogram},
+    metrics::{counter::Counter, family::Family, histogram::Histogram},
 };
 use std::sync::LazyLock;
 
@@ -44,15 +44,6 @@ pub static INVALID_TIPSET_TOTAL: LazyLock<Counter> = LazyLock::new(|| {
     );
     metric
 });
-pub static HEAD_EPOCH: LazyLock<Gauge> = LazyLock::new(|| {
-    let metric = Gauge::default();
-    crate::metrics::default_registry().register(
-        "head_epoch",
-        "Latest epoch synchronized to the node",
-        metric.clone(),
-    );
-    metric
-});
 
 #[derive(Clone, Debug, Hash, PartialEq, Eq)]
 pub struct Libp2pMessageKindLabel(&'static str);

src/daemon/mod.rs

@@ -15,8 +15,10 @@ use crate::cli_shared::{
     chain_path,
     cli::{CliOpts, Config},
 };
-use crate::daemon::context::{AppContext, DbType};
-use crate::daemon::db_util::import_chain_as_forest_car;
+use crate::daemon::{
+    context::{AppContext, DbType},
+    db_util::import_chain_as_forest_car,
+};
 use crate::db::gc::SnapshotGarbageCollector;
 use crate::db::ttl::EthMappingCollector;
 use crate::libp2p::{Libp2pService, PeerManager};
@@ -26,6 +28,7 @@ use crate::rpc::RPCState;
 use crate::rpc::eth::filter::EthEventHandler;
 use crate::rpc::start_rpc;
 use crate::shim::clock::ChainEpoch;
+use crate::shim::state_tree::StateTree;
 use crate::shim::version::NetworkVersion;
 use crate::utils;
 use crate::utils::{proofs_api::ensure_proof_params_downloaded, version::FOREST_VERSION_STRING};
@@ -202,10 +205,47 @@ async fn maybe_start_metrics_service(
         );
         let db_directory = crate::db::db_engine::db_root(&chain_path(config))?;
         let db = ctx.db.writer().clone();
-        services.spawn(async {
-            crate::metrics::init_prometheus(prometheus_listener, db_directory, db)
+
+        let get_chain_head_height = Arc::new({
+            // Use `Weak` to not dead lock GC.
+            let chain_store = Arc::downgrade(ctx.state_manager.chain_store());
+            move || {
+                chain_store
+                    .upgrade()
+                    .map(|cs| cs.heaviest_tipset().epoch())
+                    .unwrap_or_default()
+            }
+        });
+        let get_chain_head_actor_version = Arc::new({
+            // Use `Weak` to not dead lock GC.
+            let chain_store = Arc::downgrade(ctx.state_manager.chain_store());
+            move || {
+                if let Some(cs) = chain_store.upgrade()
+                    && let Ok(state) =
+                        StateTree::new_from_root(cs.db.clone(), cs.heaviest_tipset().parent_state())
+                    && let Ok(bundle_meta) = state.get_actor_bundle_metadata()
+                    && let Ok(actor_version) = bundle_meta.actor_major_version()
+                {
+                    return actor_version;
+                }
+                0
+            }
+        });
+        services.spawn({
+            let chain_config = ctx.chain_config().clone();
+            let get_chain_head_height = get_chain_head_height.clone();
+            async {
+                crate::metrics::init_prometheus(
+                    prometheus_listener,
+                    db_directory,
+                    db,
+                    chain_config,
+                    get_chain_head_height,
+                    get_chain_head_actor_version,
+                )
                 .await
                 .context("Failed to initiate prometheus server")
+            }
         });
 
         crate::metrics::register_collector(Box::new(
@@ -215,6 +255,7 @@ async fn maybe_start_metrics_service(
                     .chain_store()
                     .genesis_block_header()
                     .timestamp,
+                get_chain_head_height,
             ),
         ));
     }

src/metrics/mod.rs

@@ -3,7 +3,7 @@
 
 pub mod db;
 
-use crate::db::DBStatistics;
+use crate::{db::DBStatistics, networks::ChainConfig, shim::clock::ChainEpoch};
 use axum::{Router, http::StatusCode, response::IntoResponse, routing::get};
 use parking_lot::{RwLock, RwLockWriteGuard};
 use prometheus_client::{
@@ -86,6 +86,9 @@ pub async fn init_prometheus<DB>(
     prometheus_listener: TcpListener,
     db_directory: PathBuf,
     db: Arc<DB>,
+    chain_config: Arc<ChainConfig>,
+    get_chain_head_height: Arc<impl Fn() -> ChainEpoch + Send + Sync + 'static>,
+    get_chain_head_actor_version: Arc<impl Fn() -> u64 + Send + Sync + 'static>,
 ) -> anyhow::Result<()>
 where
     DB: DBStatistics + Send + Sync + 'static,
@@ -101,6 +104,13 @@ where
         crate::utils::version::ForestVersionCollector::new(),
     ));
     register_collector(Box::new(crate::metrics::db::DBCollector::new(db_directory)));
+    register_collector(Box::new(
+        crate::networks::metrics::NetworkVersionCollector::new(
+            chain_config,
+            get_chain_head_height,
+            get_chain_head_actor_version,
+        ),
+    ));
 
     // Create an configure HTTP server
     let app = Router::new()